Automated City Modeling from Omnidirectional Image Sequences

City models are needed in many areas of human activities including urban planning, virtual reality, and architecture preservation. Aerial, ground level, and omnidirectional city modeling techniques are compared and their main advantages and disadvantages are presented. We also show several city modeling systems and address their novelties as well as their limitations. Next, we present our previous work dealing with a step towards a 3D reconstruction system for city modeling from omnidirectional video sequences using structure from motion together with stereo constraints. We concentrate on two issues. First, we show how the tracking and reconstruction paradigm were adapted to use omnidirectional images taken by lenses with 180 degrees field of view. Secondly, we compare the results of the reconstruction using additional stereo constraints to the results when these constraints are not used. Performance of the system is demonstrated on a sequence of 870 images acquired while driving in a city. Finally, scalability and other open problems of city modeling from image sequences are presented together with problems which have been solved already. The main goal of the thesis is to design a scalable modeling and localization system consisting of two mutually interacting modules: a structure from motion (SfM) module and a near identical image detection (NIID) module.